Mixed signal. Agents can reach the page, but guess from there.

Everlane scores 59 out of 100, a mixed signal that reflects a site built for humans who read marketing copy, not agents that parse specifications. The plumbing an AI shopper needs to find and price a product is largely in place: robots.txt at everlane.com/robots.txt greets GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and six other crawlers with open arms, JSON-LD Product plus Offer markup hits nine out of nine sampled PDPs, and raw HTML prices resolve without JavaScript on every product tested. But once an agent tries to actually understand what it is buying, or transact on behalf of a user, the surface gets rough.

The most expensive miss is agent-native content, where Everlane loses nine points across four checks. Specification completeness scored 0 of 4: only one PDP out of nine surfaced concrete facts like material composition, fit, or dimensions in a form the classifier could extract, with pages like the women's city boot in sand suede returning nothing beyond "$125.00 away from free standard shipping" and "30-Day Easy Returns." Conversational descriptions scored 0 of 2. Decision-support metadata (size charts, fit guides, review breakdowns) scored 0 of 3. Delivery dates scored 0 of 2, with only two of nine pages surfacing even unstructured speed language and zero structured shipping schema. For a brand whose entire marketing identity rests on "radical transparency" about materials and factories, the absence of machine-readable material and fit data on the product pages themselves is a strategic own-goal.

The second wound is transactability, where the checkout flow scored 2 of 6 and guest checkout scored 2 of 4. Everlane's /checkout endpoint returned a 403 to the scanner and would not render without an initialized cart, which triggered a blind spot. The cart page surfaces "SIGN UP / LOG IN" but no visible "continue as guest" affordance. Combined with CAPTCHA at 0 of 4 (reCAPTCHA and hCaptcha detected on both cart and homepage, flagged as transaction-blocking), an agent attempting an end-to-end purchase runs directly into human-verification walls at exactly the wrong moment. Shop Pay, PayPal, and Apple Pay are all present, and the Shopify /cart endpoint is accessible, so the rails exist. The gate is the CAPTCHA.

Structured attributes and inventory are half-finished. Product attributes structured scored 3 of 6: seven of nine PDPs carry brand, MPN, and SKU in JSON-LD, but two (the black suede city boot and the navy dolman silk top) carry none. Inventory signals scored 2 of 5 because every page hits only the partial band, mixing schema availability with text availability but never surfacing shipping or ETA in schema. The sitemap at everlane.com/sitemap.xml contains just 9 URLs with no product entries and no lastmod timestamps, costing another 2 points.

Trust and posture round out the drag. Stated agent policy scored 0 of 2 (terms and privacy are silent on automated access), merchant verifiability scored 1 of 3 (no physical address, no registration ID), and return policy scored 1 of 3 (window present, conditions and process absent from machine-readable form).

The biggest opportunity: strip the CAPTCHAs off cart and homepage and publish the material, fit, and delivery-date data Everlane already has in structured form on the product page.